So you have two NPS RADIUS proxies on both A and B sites and use local one as primary and remote one as secodary? I didn't read that in your original post.
If that's the case, FGT-A needs to reach NPS-B which is the mirror image of FGT-B needing to reach NPS-A. So when FGT-A tries to reach NPS-B, and the route is pointing into the tunnel, FGT-A tries picking up the tunnel IP as its source. Then you want to have an IP there too.
Generally it's a good idea to assign a /30 (or /31) subnet to the tunnel and configure one IP to both sides, like 10.10.1.1/30 on A, 10.10.1.2/30 on B side. Then set phase2 selectors on both sides to allow the IP to get to/get back from the opposite NPS server IP as Debbie said.
These tunnel IPs are useful when you set specific routing toward the tunnel and when you debug with traceroute through the tunnel. Be aware FGT doesn't like the actual interface IP to be configured as /30 for the prefix-length. You need to configure them as 10.10.1.1/32 local, 10.10.1.2/32 remote on one end, and vice versa on the other end. Because they would be put in the routing table as connected routes. If you configure /30s for local and remote, both overlap in the routing table which is not allowed.